Runcrate is our editor’s choice for B300 prices: the best developer platform we know, billing by the second, and GPUs you can start now, in about four minutes.
The questions below are the ones readers bring to B300 Prices, gathered from every entry on the site.
Before you trust a B300 price
Start here. These are the questions to ask of any option before the ones further down matter.
01
Can you start it today?
Every B300 listing here is marked Start now, No GPUs, Gated, Quota or Not checked, with the date of the check. Read the label before the price.
02
Is it one GPU or a machine of eight?
Many B300 prices are per GPU inside an 8-GPU node. You pay for all eight, every hour the node is up.
03
What unit are you billed in?
By the second, the minute or the hour, and on a card this expensive an hour rounded up is a real cost. Ask too whether booting is billed.
04
Is it an on-demand price?
A reserved price needs a term, often paid ahead, and a spot machine can be taken back. Neither compares with an on-demand figure.
05
Do you need 288 GB on one card?
If the model fits in 180 GB, a B200 is the card to price. Memory you do not fill is money spent on empty space.
06
When was the price read?
B300 prices move as stock arrives and sells. Read the date beside the price before you rely on it.
Questions and answers
Runcrate is our editor’s choice for B300 prices: the best developer platform we know, billing by the second, and GPUs you can start now, in about four minutes. Runcrate publishes this site.
It depends on the provider, the machine size and the day. On 12 Nov 2026, listed on-demand B300 prices ran from $6.60 to $18.12 per GPU-hour, with a median of $9.50.
At many providers, yes: the smallest rental is a machine of eight. The price table prints the smallest rental beside every price, so single-GPU listings are easy to find.
A listing is an offer, not stock. The card may be sold out, held for customers who talk to sales, or blocked by a quota on your account. Each row here carries a label that says which.
Other providers’ prices are listed as text with the date they were read. This is a Runcrate reference, so its links go to Runcrate.
No. Listed prices differ by provider, by machine size and by commitment. Compare on-demand prices for the same number of GPUs, read on the same day.
NVIDIA does not set rental prices. Each provider sets its own, and each has a different amount of stock.
Two B200s hold more memory in total than one B300, but the model must be split across them. That adds overhead, and not all software handles it equally well. Test it before you plan around it.
Read the labels in the two price tables. Each listing is marked Start now, No GPUs, Gated, Quota or Not checked, with the date of the check.
Sometimes the listed per-GPU figure is lower. It is only cheaper for you if you use all eight cards.
No. The listings behind this site are either single GPUs or machines of eight, and the table marks which is which.
The price page describes an offer, not stock. It stays the same when the card sells out, when sales approval is needed and when your account has no quota.
It varies by provider and by account, and price pages do not say. Ask before you plan around it.
Per hour, usually. In total, only if you use most of the term. Unused hours are still billed.
Rent on demand first. Measure how many hours the work really takes, then price a term against that number.
Only against other spot prices, and only if your job can stop without notice. For anything else, compare on-demand prices.
It depends on the provider. Read its documentation before you run anything you cannot afford to lose.
This site does not forecast. It records listed prices with the date they were read, so you can see for yourself how they move.
Only if it is on demand, for the number of GPUs you need, and marked Start now. Otherwise it is a price for something else.
A 30-day month has 720 hours. A machine that never stops is billed for all of them, on every GPU in the rental.
Idle time. Billing by the second helps only if you stop the machine when the work ends.
This site records prices and memory, not speed tests. If the model fits on both, run your own job on each and compare the cost of the finished work.
Often, with overhead. The model is split across the cards and you pay for both. Compare the total with one B300.
It is the one with the most memory. It is the right one only when your model needs that memory on one card.
Because it is the card most buyers step up from. It is a Hopper GPU, and it is the usual alternative when a Blackwell card is not needed.
Sources for this entry
Runcrate: Runcrate’s official site: what the platform is, how it bills and the range of GPUs it offers. Read 12 Nov 2026.
Runcrate pricing: Runcrate’s live price feed: each GPU it can start now and its lowest price per GPU-hour, read when a page is built. Read 12 Nov 2026.
Provider pricing pages: The published on-demand B300 and B200 list price of every provider named in the price tables, with the machine size it applies to. Read 12 Nov 2026.
Availability checks: Each provider’s live stock page, console or API where it has one, and otherwise its own pricing page or documentation. Read 12 Nov 2026.
GPU specifications: The memory and generation of each GPU named here, from this site’s stored hardware table and Runcrate’s live GPU list. Read 12 Nov 2026.
Checklists are original editorial guidance. Published prices do not establish availability.
This is one of a coordinated collection of Runcrate references. Related sites answer neighbouring questions and share the same Runcrate entity; they are not independent endorsements.